Inefficiency of __getattr__

[Huaiyu Zhu]

Is __getattr__ supposed to be run only when the attribute is not already
defined?  This does not seem to be so and it is very costly (equivalent of
addition of two 1000 element vectors).  In the following example __getattr__
should not be invoked even if it is there:

======================== test program ===========================
class A:
    def __add__(self, other):   pass
    def __getattr__(self, name):
        try:                return self.__dict__[name]
        except KeyError:    raise AttributeError, name

a = A()
def test():
    for i in xrange(10000):     a+a

from profile import run
run("test()")

==================================================================
Result with __getattr__ defined:

         20003 function calls (10004 primitive calls) in 1.870 CPU seconds

   Ordered by: standard name

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
        1    0.000    0.000    1.870    1.870 <string>:1(?)
        0    0.000             0.000          profile:0(profiler)
        1    0.000    0.000    1.870    1.870 profile:0(test())
    10000    0.220    0.000    0.220    0.000 test_getattr.py:2(__add__)
  10000/1    1.650    0.000    1.870    1.870 test_getattr.py:3(__getattr__)
        1    0.000    0.000    1.870    1.870 test_getattr.py:8(test)

==================================================================
Result with __getattr__ commented out:

         10003 function calls in 0.560 CPU seconds

   Ordered by: standard name

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
        1    0.000    0.000    0.530    0.530 <string>:1(?)
        0    0.000             0.000          profile:0(profiler)
        1    0.030    0.030    0.560    0.560 profile:0(test())
    10000    0.170    0.000    0.170    0.000 test_getattr.py:2(__add__)
        1    0.360    0.360    0.530    0.530 test_getattr.py:8(test)

-- 
Huaiyu Zhu                       hzhu at users.sourceforge.net